Published works

The propagation of bushfire across the landscape is dependent on a variety of environmental factors, but the wind, in particular, has a major effect on both the speed and direction of fire propagation. As such, bushfire spread models, which underpin successful bushfire management, require accurate knowledge of the pattern of winds across the landscape. This can be problematic over complex terrain where winds exhibit considerable spatial variability due to wind-terrain interactions, and where detailed measurements of wind characteristics are comparatively rare.

This thesis contributes two new wind datasets to address the previous lack of data available to develop and validate wind models over complex terrain. It also details analyses that focus on the statistical characterisation of wind as joint wind direction distributions, which represent the directional wind response to changing topography and surface roughness. A novel method for toroidal surface fitting is introduced and implemented to estimate the true continuous response from discrete observed data. This new method, which relies on a conceptually simple adaptation of planar techniques, is compared to the limited range of available toroidal surface estimation techniques and is shown to perform as well as, if not better than, these more sophisticated methods.

Monte Carlo simulations are employed to highlight the sensitivity of statistical comparison tests to alternative distribution structures, and to validate bivariate and circular extensions of the Kolmogorov-Smirnov test. These tests are applied to directional wind response pairs, showing that vegetation regrowth has a significant but varying impact across complex terrain.

Finally, this thesis demonstrates how statistical approaches can be used to complement current physics-based wind modelling methods. The resulting probabilistic representations provide more accurate predictions of wind direction variability, and are better suited to emerging ensemble-based bushfire prediction frameworks. As such, they provide a superior characterisation of uncertainty across the fire modelling process; ultimately enabling fire managers to make more informed decisions.